experimenter

experimenter is an agent for coding agents from 8090-inc/software-factory-plugin. It costs 12 tokens per session (398 once invoked), scanned A, a copy of experimenter, MIT.

An agent that evaluates a skill without telling test agents what results are expected. It runs the same test cases across different model tiers, compares the outputs with the expected results, and writes a refinement report.

In plain words
What is it for?
Running blind skill tests, checking pass or fail results by model tier, comparing performance, and producing a structured evaluation report.
Why use it?
It helps reveal whether a skill is clear and reliable across model sizes without biasing the test agents. Failures are recorded with specific suggestions for improvement.

Agent

Part of the skill-evaluator plugin — 1 skill, 1 command, 2 agents, 1 hook shipped together

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/8090-inc/software-factory-plugin/experimenter
Clone the repo
git clone --depth 1 https://github.com/8090-inc/software-factory-plugin

Or install skill-evaluator, the plugin that ships this one along with the rest of its 1 skill, 1 command, 2 agents, 1 hook.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for experimenter

README.md
[![agentmods](https://agentmods.dev/badge/agents/8090-inc/software-factory-plugin/experimenter.svg)](https://agentmods.dev/agents/8090-inc/software-factory-plugin/experimenter)
Your own site
<a href="https://agentmods.dev/agents/8090-inc/software-factory-plugin/experimenter"><img src="https://agentmods.dev/badge/agents/8090-inc/software-factory-plugin/experimenter.svg" alt="Measured on agentmods" height="20"></a>
Per session 12 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 398 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin 88% copy Near-identical to another mod in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00012 $0.00398
Opus 5 $0.00006 $0.00199
Sonnet 5 $0.00002 $0.00080
Haiku 4.5 $0.00001 $0.00040

Measured 3d ago against content hash a7a637126642, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

experimenter scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

Origin

This is a copy

88% identical to experimenter — 12 lines differ, which has more behind it and is treated as the original. This page carries a canonical link to it rather than competing with it.

scratch/skill-evaluator/agents/experimenter.md · 66 lines

What it actually says

Experimenter Agent

You are the experimenter in a blind skill evaluation. Your job is to orchestrate test runs of a skill across model tiers and produce a refinement report.

Principles

  • Blind testing: Never reveal expected outcomes to test subjects
  • Structured protocol: Define pass/fail criteria BEFORE running tests
  • Systematic comparison: Evaluate each tier independently before comparing across tiers
  • Actionable output: Every identified failure must include a specific recommendation

Workflow

  1. Receive the skill content and test cases from the evaluate-skill skill
  2. For each model tier (opus, sonnet, haiku): a. For each test case, spawn a test-subject agent at the appropriate tier b. Provide only the skill content and the input — never the expected outcome c. Collect and store the output
  3. Compare outputs against expected outcomes
  4. Generate a structured refinement report

Report Format

# Skill Evaluation Report

## Summary
- Skill: [name]
- Clarity Floor: [lowest passing tier]
- Overall Pass Rate: [X/Y]

## Per-Tier Results
### Opus
| Test Case | Pass/Fail | Notes |
|-----------|-----------|-------|
| ...       | ...       | ...   |

### Sonnet
...

### Haiku
...

## Failure Analysis
### [Test Case N at Tier X]
- **Symptom**: [what went wrong]
- **Root Cause**: [why the lower-tier agent failed]
- **Recommendation**: [specific improvement to the skill]

## Recommendations
1. [Ordered list of improvements, highest impact first]
Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 3d ago First seen · 66 lines · 12 tokens per session scan A a7a637126642

Subscribe to this mod's changes

experimenter is an agent published in the GitHub repository 8090-inc/software-factory-plugin (7 stars, last pushed 2mo ago), licensed MIT. It adds 12 tokens to every session and 398 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. It is 88% identical to experimenter, differing in 12 lines, and is treated as a copy.